具有噪声干扰的非线性时变系统多维泰勒网辨识和控制
Multi-dimensional Taylor network identification and control of nonlinear time-varying systems with noise disturbances
摘要点击 209  全文点击 194  投稿时间:2018-06-21  修订日期:2019-04-21
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2019.80462
  2020,37(1):107-117
中文关键词  多维泰勒网  非线性时变系统  噪声干扰  自适应控制  非线性滤波
英文关键词  Multi-dimensional Taylor network  nonlinear time-varying system  noise disturbance  adaptive control  nonlinear filtering
基金项目  
作者单位E-mail
张超 河南工学院 zhangchao915@foxmail.com 
孙启鸣 南京林业大学  
中文摘要
      针对含噪声单输入单输出不确定非线性时变系统,提出一种基于多维泰勒网(MTN)稳定的自适应控制方案,其中三个MTN分别被用来实现非线性滤波、系统辨识与自适应控制。首先,MTN滤波器(MTNF)用来消除测量噪声,以得到无随机干扰的模型输出。然后,MTN辨识器(MTNI)用来表示系统动态映射且比传统神经网络泛化能力更强。而后,MTN控制器(MTNC)用来实现系统精确跟踪控制,其中时变被控对象由MTNI辨识并将其动力学特性信息实时提供给MTNC使其“光滑”自适应。此外,利用改进的灵敏度计算方法来剪除MTNI和MTNC的冗余输入和冗余中间层回归项。最后,证明基于MTN的闭环系统稳定性,并给出最优学习率以期实现快速学习。仿真结果表明,该方法具有精确的辨识能力、良好的跟踪性能和较强的抗干扰能力,可实现含有不确定性、随机因素和时变特性的非线性系统自适应实时控制。
英文摘要
      For the uncertain single-input single-output no-linear time-varying system with noise, an adaptive stable control scheme based on multi-dimensional Taylor network (MTN) is proposed. Three of the MTNs are used respectively to implement nonlinear filtering, system identification and adaptive control. First of all, the MTN filter (MTNF) is used to eliminate measurement noise to obtain model output without random interference. Then, the MTNI identifier (MTNI), with better generalization ability than the traditional neural network, is used to represent the dynamic mapping of the system. After that, the MTN controller (MTNC) is used to achieve accurate tracking control of the system, in which the time-varying controlled object is identified by MTNI and its dynamic characteristic information is provided to MTNC in real time to make it "smooth" adaptive. In addition, an improved sensitivity calculation method is used to trim the redundant input and redundant middle layer regressive terms of MTNI and MTNC. Finally, the stability of the closed-loop system based on MTN is proved and the optimal learning rate is provided to achieve rapid learning. Simulation results show that the proposed method, which has accurate identification capability, good tracking performance and strong anti-jamming capability, can realize adaptive real-time control of nonlinear systems with uncertainties, random factors and time-varying characteristics.